Agentic systems can load tools, check web content and ask clarifying questions. GPT-6 Astra and GPT-5.6 Sol illustrate different workflows in this article; we do not infer a general performance ranking from them. For Swiss SMEs, the relevant question is whether the website and process support the actual task.
What GPT-6 Astra changes
The graphic contrasts a continuous task with a workflow involving tools and questions. It is an editorial illustration, not a published comparative test. The workflow also depends on tool access, product interface, settings and task. Without documented versions and equivalent test conditions, a cross-model performance claim is not substantiated.
That is not a cosmetic UI update. It changes who reads in the buying process. A person clicks a listing. An agent is supposed to finish an action: compare, book, request a proposal, map a local service. Copy written only for the click is missing in that second pass.
Vendor comparisons claim large jumps on computer-use tasks. GlasBox does not take those scores as a client KPI. We measure what sits on your domain and in visible answers — not a demo score.
SEO, GEO and a separate RAG commission
Listings and generated answers can share a search foundation. RAG is a retrieval technique for public or internal sources. GlasBox’s internal RAG offering is a separate knowledge system with roles, approved sources and its own acceptance tests.
Crawl, index, canonical URLs, internal links, load time and service pages in human language. Without that base no agent has anything reliable to read.
The work that helps generative search systems — ChatGPT, Gemini, Perplexity, Google AI Overviews — place your company correctly and, when the evidence exists, treat you as a source. GEO does not replace SEO. It sits on top of it.
RAG retrieves public or internal information as answer context. An internal knowledge system additionally requires access, source and escalation checks. A cited website does not replace that work.
GEO: making search AI treat your company as a source
Public AI search and internal knowledge are two systems
Agent recommendation, not only the click
For an agentic task, we recommend consistent company names, addresses and service terms, together with working enquiry paths. These properties can be checked on the website. Whether an agent produces a recommendation is observed in the actual workflow, not promised as a consequence.
When an agent should request a proposal, find a slot or map a local service, it fails on ambiguous names, dead forms and pages that rename the same offer in every paragraph. That is not new magic. It is information architecture a model can extract.
3D worlds, UE5 scenes or immersive demos can be extra signals. They do not replace a crawlable service page, an imprint or a consistent entity. GlasBox does not sell that as a separate switch or as a guarantee of agent recommendations.
What makes a site usable for agents and answer engines
Search-relevant pages are accessible and indexable. Search retrieval and model training are controlled separately using each provider’s rules.
Entity signals match: GlasBox Studio, Ruta Group, Werkstrasse 11, 6102 Malters, LU, Switzerland — and on your side the same name, address and services.
Each section answers a decision: what it is, for whom, what it is not, what the next checkable step costs.
Forms are reachable and specific: proposal, short check, audit — not a generic “contact” with no intent.
Public facts and internal knowledge stay apart. RAG is not sold as an SEO trick.
SEO, GEO and AI visibility
Controlled RAG systems
Client acquisition: measure, do not demo-wash
An agentic model can fill spreadsheets and build dashboards. That replaces neither a brief nor a contract. GlasBox does not automate sales through a public “agent readiness” simulator and does not invent a guaranteed recommendation rate.
What we do: keep the free AI-visibility short check limited to an initial assessment — one domain, three customer questions, three observations. SOURCE/01 Full records SEO, AI visibility and knowledge fitness for CHF 2'450 one-time, with no mention guarantee. The SEO/GEO mandate starts after the audit from CHF 1'500 per month. Every individually prepared project proposal includes a free preliminary analysis based on your brief and public information. It is not a full audit.
SOURCE/01 overview
What this article does not claim
- No ranking, mention or lead guarantee from GPT-6 Astra, ChatGPT or Google.
- No unreviewed vendor benchmarks as a project target.
- No public fixed RAG price. RAG stays suitability plus proposal.
- No model replaces nDSG, hosting choice or human approval.
An acceptance task for an agentic workflow
Example acceptance test: a system must find the scope of a service and prepare an enquiry draft. Check the source used, accuracy and handover to the human. Submitting a binding request is a separate step requiring clear approval.
| Step | Check | Acceptance |
|---|---|---|
| Find the service | Canonical page and visible conditions | Text agrees with the source. |
| Prepare the enquiry | Required fields, request and contact details | No invented information. |
| Handover | Show draft and unresolved questions | A human can review and approve. |
Next step
GlasBox measures classic SEO and AI visibility separately. No ranking or mention guarantee. The preliminary analysis in the proposal is free and does not replace SOURCE/01 Full.
Request a proposal with free preliminary analysis